fetch

Official MCP reference server that fetches web pages and converts them to markdown for the model. Can reach internal IP addresses, so sandbox it.

91.2k ★(monorepo)+161/wkMCPestablishedApache-2.0 / MIT (transitioning)

Score

85/100

91,237 stars

10,424 forks

A low score is not a verdict on quality. Young and niche tools start low by design. How we calculate scores

The Open Source Drop Lens

By Erik Loyd, SaaS CEO and former COO/CFO of an AWS Premier Partner.

Updated Sep 2026

Fetch hands your AI assistant a URL reader. The model asks for a page, and the server returns it converted to markdown, in chunks the model can page through when the content runs long.

Setup is one uvx mcp-server-fetch entry in your config, or a Docker image. It obeys robots.txt for model-initiated requests unless you pass --ignore-robots-txt, and it takes a custom user agent and a proxy. No API keys, no service fees.

The MCP steering group maintains it as a reference implementation, and its README carries a caution worth reading twice: it can reach local and internal IP addresses. Fine on a laptop. On any network with internal services, sandbox it first.

Cost Breakdown

fully free

Free tier: Free and open source (the project is moving from MIT to Apache 2.0). No API keys and no paid service behind it.

Self-hosted: Runs locally with uvx, pip or Docker. Installing Node.js is optional and switches it to a more robust HTML simplifier.

Paid: None. The only cost is exposure: the server can request local and internal IP addresses, so treat it as a network-capable process.

No service fees. The real cost is risk: it can fetch internal IP addresses, so sandbox it on any network with private services.

Get tools like this every Wednesday

One featured tool, three on the radar. No fluff.

Explore Further

More tools in the directory